@manya-hael/memory

@manya/memory

A true memory engine — not just a vector database.

@manya/memory composes three existing Manya tools into five distinct memory types, so different intelligences in the ecosystem can share memory without sharing everything.

Sharing across intelligences happens over @manya/unify's event bus, and only on channels you explicitly publish to with shareOn. Nothing leaves a memory store implicitly.

Usage

import { memory } from '@manya/memory';

const store = memory.createMemoryStore('ara');

// Episodic — what happened
memory.remember(store, 'ara', 'noticed the front door unlocked');
memory.verifyEpisodicIntegrity(store); // { valid: true, ... }

// Working — short-lived context
memory.rememberWorking(store, 'current-task', { step: 'confirm with owner' }, 60_000);

// Semantic — learned facts
memory.learnFact(store, 'owner', 'prefers texts over calls after 9pm', 0.95);

// Procedural — callable skills
memory.learnSkill(store, 'summarize', (text) => text.slice(0, 140));

// Archival — encrypted long-term storage
memory.archive(store, 'family-rules', { quietHours: '21:00-07:00' });

// Selective sharing — only subscribers of this exact channel see it
memory.subscribeOn(store, 'household-updates', (evt) => console.log(evt.payload));
memory.shareOn(store, 'household-updates', { note: 'door was unlocked, now locked' });

Why this composition

• episodic → stamp — every event is chained to the previous one, so the history can't be silently rewritten. verifyEpisodicIntegrity catches it if it is.
• archival → vault — anything meant to outlive a session goes through vault's encrypted store, the same one identity and secrets already use.
• sharing → unify — reuses the same event bus the rest of the mesh runs on, instead of inventing a second pub/sub mechanism.